基于神经元网络的选择型问卷识别技术

在问卷调查数据自动识别和统计过程中,由于纸张的折叠、弯曲、变形及受污染等原因引起的数据误判时有发生,考虑到手工统计的繁杂性,开发一种自动、高效的智能处理系统具有相当大的实用价值。提出应用神经网络对调查问卷扫描图像进行识别处理的方法,建立基于MATLAB的Hopfield网络识别模型,并详细讨论了图像预处理、特征提取及Hopfield网络训练与识别这三个重要环节。针对建立好的识别模型,系统仿真情况下,符号识别率达到100%;在实际操作过程中,当训练样本数充足,样本来源可靠的情况下,识别率高达96%,基本实现预期效果。

Technology of Selective Questionnaire Recognition based on Neural Network

In the process of automatic questionnaire recognition and statistic, there always is data misjudgment caused by folding, bending, distorting and contamination of the paper. Considering complexity in manual statistic, it is valuable for developing an intelligent process system with automation and high efficiency. This paper puts forward data process method based on neural network to handle the scanned image of questionnaire, sets up the Hopfield network recognition model based on MATLAB. Then it discusses specifically about three important parts of this procedure, such as image preprocessing, feature extraction, Hopfield network training and recognition. According to the established model, system simulation could get 100% of recognition rate. In real operation, once the sufficient and reliable sampies are obtained, recognition rate could reach 96%, so the desired effect are achieved.